{
 "chart": "baseline_results/baseline_results",
 "title": "Public frontier models outperformed every human baseliner on three tasks we baselined",
 "subtitle": "Hours are the best baseliner's total time on the task, not the time they took to reach their top score. Models had their own task-specific time and compute limits.",
 "panels": [
  {
   "title": "",
   "x": {"label": "Score", "unit": "", "scale": "linear"},
   "y": {"label": "", "unit": "", "scale": "linear"},
   "tables": [
    {
     "name": "tasks",
     "rows": [
      {"task": "Reverse Conway", "baseliners": 2, "best_human_score": 0.137, "best_human_hours": 39.0, "best_model_score": 0.208405},
      {"task": "Minecraft Bot", "baseliners": 1, "best_human_score": 0.055, "best_human_hours": 160.0, "best_model_score": 0.175},
      {"task": "Train a Program", "baseliners": 4, "best_human_score": 0.042, "best_human_hours": 45.0, "best_model_score": 0.44334}
     ]
    }
   ]
  }
 ]
}
